Choosing Numerical Methods for Derivative Pricing
Summary
The answer gives a broad guide to selecting pricing methods by payoff and model structure. For vanilla European options and digitals, it describes the Black–Scholes formula as a practical way to interpolate implied volatility from traded options, while noting that extrapolation beyond observed market quotes creates greater risk. Quanto European pricing is also mentioned, with implicit assumptions about the terminal distributions of the underlying and exchange rate, plus a correlation input that may require additional modeling.
For claims depending only on the underlying’s terminal distribution, the answer points to numerical integration. For path-dependent payoffs with a single underlying and low dimensionality, it suggests finite-difference PDE methods or trees. Monte Carlo is presented as better suited to path-dependent, higher-dimensional products such as baskets or interest-rate products with multiple forward rates. These are rules of thumb rather than a full comparison: the answer does not detail convergence, computational cost, calibration, or specific approaches for Lévy models, stochastic volatility, or American exercise.
Key ideas
- Black–Scholes can be used to interpolate implied volatility for vanilla European options and digitals from traded prices.
- Extrapolating an implied volatility surface is identified as a source of greater uncertainty.
- Numerical integration is suggested for payoffs that depend only on the terminal distribution.
- Finite-difference PDE methods or trees suit low-dimensional, path-dependent payoffs with a single underlying.
- Monte Carlo is suggested for path-dependent products with multiple risk factors or underlyings.
Tags
Full text
# What are the industry standards and rules of thumb when it comes to numerical methods? # What are the industry standards and rules of thumb when it comes to numerical methods? So, as far as I know, we have 3 main numerical methods. Monte Carlo, PDE-methods (FDM), and numerical integration methods (Fourier transforms and so on). How do these methods generally compare to each other? What are the rules of thumb? When should you use one over the other? I am not asking for specifics, just general ideas for the most typical situations that one might run into. For example, for European option pricing in Levy models, which are most often used, and why? What if we switch to stochastic volatility models? What if we want to price Americans or path-dependents? ## Answer by bhutes (score 4) https://quant.stackexchange.com/a/46523 1> Analytical - Black Scholes formula for Vanilla European options, Digitals. These valuations are just an "interpolation" of traded options. We interpolate the implied volatility from the traded points on the implied volatility surface. There is no modeling assumption involved here. Market uses this formula for implementing the "interpolation". Scope for going wrong using this method (compared to market participants) is very limited. Most issues arise only when volatility surface needs to be extrapolated. Quanto Europeans also use this method frequently, although this is more than just an interpolation. Modeling assumptions are implicit - (i) Lognormal terminal distributions of the underlying and the FX rate, (ii) correlation input to the analytical formula is often not an "interpolated" value from traded options; hence needs further modeling assumptions. 2> Numerical integration - all payoffs which depend only on the terminal distribution of the underlying, e.g. self-quanto options 3> PDE Method (Finite difference or Trees) - for path dependent payoffs with single underliers (low dimensionality). 4> Monte Carlo - for path dependent payoffs with higher dimensionality e.g. basket options or rates products with multiple Forward Libor rates as the underliers.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.